[S01] PM-1

[S01-14] On the sensitivity kernels for Rayleigh wave azimuthal anisotropy

*Hitoshi KAWAKATSU^1,2 (1. Earthquake Recearch Institute, 2. Institute of Earth Sciences, Academia Sinica)

For the investigation of the oceanic lithosphere-asthenosphere system employing regional OBS
array data, it is essential to resolve azimuthal anisotropy as well as radial anisotropy (e.g., Takeo et al.,
2018). Considering the presence of strong sub-Moho anisotropy both in P_n and S_n, it is important
to analyze both P and S anisotropy for Rayleigh waves as the phase velocity is quite sensitive to
the shallow-most P-wave structure. It is well established that, in the case of weak anisotropy, the
sensitivity kernels of azimuthal anisotropy have the same form as those for the corresponding VTI
parameters (Montagner and Nataf, 1986). This may be understood as that, within the framework of
the first order perturbation theory, the eigenfunction is unchanged and thus the phase velocity change
is expressed in terms of the equivalent VTI parameters within the plane of the wave propagation.
Recently, Russell (2021, Ph.D. Thesis, Appendix E, F) discussed how parameterizations using Love
parameters (δA(B), δL(G), δN(E), δF(H)) due to (Montagner and Nataf, 1986) and those using
velocity perturbation (δα_H, δβ_V , δβ_H, δη) due to (Takeuchi and Saito, 1972) might be related for
azimuthal anisotropy. After pointing out that solving for δβ_V/β_V (or δα_H/α_H) on its own is not
equivalent to solving for G/L (or B/A) because of the presence of the δF-term, they suggest assuming δη/η∼0 for inversion. However, the assumption δη/η=0 is exactly the reason why the usage of the
conventional η introduces the unpreferable behavior of P-wave kernels that is contaminated by the
S-wave sensitivities (Kawakatsu, 2016) and unjustifiable; also existing mantle fabrics show a strong
azimuthal dependence for δη/η. Instead, we suggest using the parameterization involving the new fifth parameter η_κ, i.e, (δα_H, δβ_V , δβ_H, δη_κ). Because the fifth parameter is difficult to constrain,
it might be more practical to omit the corresponding term in the inversion, which is equivalent to
assume δη_κ/η_κ∼0, meaning azimuthal independency of the "ellipticity" parameter η_κ that offers a
physical background of the modeling. Also, the aforementioned fabrics show a much weaker azimuthal
dependency for δη_κ/η_κ compared to that for δη/η. This approach may be useful for the scaling of S-
and P-wave azimuthal anisotropy to reduce the number of parameters in the Rayleigh wave inversion.